2. Data and Methods
Deep SOLO floats measure pressure and time during descent nominally at 5 to 50 dbar intervals, depending on how engineering parameters are set for an individual float. These measurements are telemetered to shore as part of the engineering data when the float surfaces during each nominally 10-day cycle, and reported in Argo trajectory (.traj) files. We only use these data from profiles that sample to at least 2995 dbar, and limit our analysis of the data to pressures greater than 800 dbar, below the main pycnocline in the regions analyzed. We also remove three profiles with vertical gaps in these engineering data that are too long to smooth over. The screening results in 10070 profiles in total, mostly in the Brazil Basin of the western South Atlantic and the Southwest Pacific Basin, but also in the South Australian Basin, the Australian-Antarctic Basin, the North American Basin (western north Atlantic), and a few in the Central and Northeast Pacific basins (Figure 1).
For each profile extending to at least 2995 dbar we estimate vertical velocities from first differences of pressures divided by those of times. We discard the three deepest estimates to avoid bottom interactions, then apply a loess smoother (Cleveland & Devlin, 1988) with a half-power point of ~150 dbar to estimate vertical velocities on a regular grid at 10-dbar intervals from 1000 dbar to the deepest pressure (rounded down to the closest 10 dbar value shallower than that pressure). We then fit a second-order polynomial to these gridded vertical velocities as an estimate of the background fall rate that steadily decreases with increasing pressure, and subtract those background fall rates from each profile to yield the residual vertical velocities (hereafter w’ ) relative to the background fall rates. Typical background fall rates are about 0.17 (±0.01) dbar s-1 at 1000 dbar, and decrease nearly linearly to about 0.07 (±0.01) dbar s-1 at profile maximum pressures, where the uncertainties (here and throughout the manuscript) are one standard deviation.
We calculate the mean deep w’ variances averaged from 1000 dbar to the maximum pressure analyzed for each profile (Figure 2a). We also apply a Morlet wavelet transform (Torrence & Compo, 1998) to each profile to estimate the dominant vertical wavelength of w’(Figure 2b). We define this dominant vertical wavelength as the location of the maximum value of the wavelet power spectrum of w’ for all vertical wavelengths and pressures not influenced by zero-padding edge effects (i.e., heeding the “cone of influence”).
We look at the correlation of the mean deep w’ variances and dominant vertical wavelengths with local bottom topographic roughness. To do that we compute the variance of ocean bathymetry (GEBCO Compilation Group, 2021) at 1’ resolution within 40-km square bins on a 0.25° x 0.25° grid and use the resulting map to interpolate roughness to float locations.